Delve

wget http://fantom.gsc.riken.jp/5/suppl/delve/delve.tgz

Delve estimates an error probability for each mapping, and by discarding those with error prob > 0.01, tags mapping equally to two places are discarded. This means that up to 1% of the mappings could be wrong and though that may seem like a low percentage, 1% of a billion, is a lot of reads.

Initially it finds a set of putative hits for virtually all reads. In practice there is no error limit and the mapping rate is usually 99.9 - 100%. For promoter reads, the 5' hit will be recovered alongside alternative mappings somewhere else in the genome. Using these initial hits a pair HMM model is derived including probabilities of matches, mismatches and gaps along the read using the Baum-Welch procedure (special case of EM).

Based on the model the full probability calculated using the forward algorithm is assigned to all putative hits. Lastly, the posterior probability of easy hit is derived by applying the Bayesian formula.